Mallet是Umass大牛开发的一个关于统计自然语言处理的l的开源库,很好的一个东西。可以用来学topic model,训练ME模型等。对于开发者来说,其官网的技术文档是非常有效的。
mallet下载地址,浏览开发者文档,只需点击相应的“Developer's Guide”。
下面以开发一个简单的最大熵分类模型为例,可参考文档。
首先下载mallet工具包,该工具包中包含代码和jar包,简单起见,我们导入mallet-2.0.7\dist下的mallet.jar和mallet-deps.jar,导入jar包过程为:项目右击->Properties->Java Build Path->Libraries,点击“Add JARs”,在路径中选取相应的jar包即可。
新建Maxent类,代码如下:
import java.io.File; import java.io.FileInputStream; import java.io.FileNotFoundException; import java.io.FileOutputStream; import java.io.FileReader; import java.io.IOException; import java.io.ObjectInputStream; import java.io.ObjectOutputStream; import java.io.Serializable; import java.util.ArrayList; import java.util.Arrays; import java.util.List; import cc.mallet.classify.Classifier; import cc.mallet.classify.ClassifierTrainer; import cc.mallet.classify.MaxEntTrainer; import cc.mallet.classify.Trial; import cc.mallet.pipe.iterator.CsvIterator; import cc.mallet.types.Alphabet; import cc.mallet.types.FeatureVector; import cc.mallet.types.Instance; import cc.mallet.types.InstanceList; import cc.mallet.types.Label; import cc.mallet.types.LabelAlphabet; import cc.mallet.types.Labeling; import cc.mallet.util.Randoms; public class Maxent implements Serializable{ //Train a classifier public static Classifier trainClassifier(InstanceList trainingInstances) { // Here we use a maximum entropy (ie polytomous logistic regression) classifier. ClassifierTrainer trainer = new MaxEntTrainer(); return trainer.train(trainingInstances); } //save a trained classifier/write a trained classifier to disk public void saveClassifier(Classifier classifier,String savePath) throws IOException{ ObjectOutputStream oos=new ObjectOutputStream(new FileOutputStream(savePath)); oos.writeObject(classifier); oos.flush(); oos.close(); } //restore a saved classifier public Classifier loadClassifier(String savedPath) throws FileNotFoundException, IOException, ClassNotFoundException{ // Here we load a serialized classifier from a file. Classifier classifier; ObjectInputStream ois = new ObjectInputStream (new FileInputStream (new File(savedPath))); classifier = (Classifier) ois.readObject(); ois.close(); return classifier; } //predict & evaluate public String predict(Classifier classifier,Instance testInstance){ Labeling labeling = classifier.classify(testInstance).getLabeling(); Label label = labeling.getBestLabel(); return (String)label.getEntry(); } public void evaluate(Classifier classifier, String testFilePath) throws IOException { InstanceList testInstances = new InstanceList(classifier.getInstancePipe()); //format of input data:[name] [label] [data ... ] CsvIterator reader = new CsvIterator(new FileReader(new File(testFilePath)),"(\\w+)\\s+(\\w+)\\s+(.*)",3, 2, 1); // (data, label, name) field indices // Add all instances loaded by the iterator to our instance list testInstances.addThruPipe(reader); Trial trial = new Trial(classifier, testInstances); //evaluation metrics.precision, recall, and F1 System.out.println("Accuracy: " + trial.getAccuracy()); System.out.println("F1 for class 'good': " + trial.getF1("good")); System.out.println("Precision for class '" + classifier.getLabelAlphabet().lookupLabel(1) + "': " + trial.getPrecision(1)); } //perform n-fold cross validation public static Trial testTrainSplit(MaxEntTrainer trainer, InstanceList instances) { int TRAINING = 0; int TESTING = 1; int VALIDATION = 2; // Split the input list into training (90%) and testing (10%) lists. InstanceList[] instanceLists = instances.split(new Randoms(), new double[] {0.9, 0.1, 0.0}); Classifier classifier = trainClassifier(instanceLists[TRAINING]); return new Trial(classifier, instanceLists[TESTING]); } public static void main(String[] args) throws FileNotFoundException,IOException{ //define training samples Alphabet featureAlphabet = new Alphabet();//特征词典 LabelAlphabet targetAlphabet = new LabelAlphabet();//类标词典 targetAlphabet.lookupIndex("positive"); targetAlphabet.lookupIndex("negative"); targetAlphabet.lookupIndex("neutral"); targetAlphabet.stopGrowth(); featureAlphabet.lookupIndex("f1"); featureAlphabet.lookupIndex("f2"); featureAlphabet.lookupIndex("f3"); InstanceList trainingInstances = new InstanceList (featureAlphabet,targetAlphabet);//实例集对象 final int size = targetAlphabet.size(); double[] featureValues1 = {1.0, 0.0, 0.0}; double[] featureValues2 = {2.0, 0.0, 0.0}; double[] featureValues3 = {0.0, 1.0, 0.0}; double[] featureValues4 = {0.0, 0.0, 1.0}; double[] featureValues5 = {0.0, 0.0, 3.0}; String[] targetValue = {"positive","positive","neutral","negative","negative"}; List<double[]> featureValues = Arrays.asList(featureValues1,featureValues2,featureValues3,featureValues4,featureValues5); int i = 0; for(double[]featureValue:featureValues){ FeatureVector featureVector = new FeatureVector(featureAlphabet, (String[])targetAlphabet.toArray(new String[size]),featureValue);//change list to array Instance instance = new Instance (featureVector,targetAlphabet.lookupLabel(targetValue[i]), "xxx",null); i++; trainingInstances.add(instance); } Maxent maxent = new Maxent(); Classifier maxentclassifier = maxent.trainClassifier(trainingInstances); //loading test examples double[] testfeatureValues = {0.5, 0.5, 6.0}; FeatureVector testfeatureVector = new FeatureVector(featureAlphabet, (String[])targetAlphabet.toArray(new String[size]),testfeatureValues); //new instance(data,target,name,source) Instance testinstance = new Instance (testfeatureVector,targetAlphabet.lookupLabel("negative"), "xxx",null); System.out.print(maxent.predict(maxentclassifier, testinstance)); //maxent.evaluate(maxentclassifier, "resource/testdata.txt"); } }
说明:trainingInstances为训练样本,testinstance为测试样本,该程序的执行结果为“negative”。